Ever wondered how your favorite apps handle millions of requests per second without breaking a sweat? In this deep dive into AWS Elastic Load Balancing, we'll explore the impressive statistics behind the scenes, from the million-request-per-second throughput of Application Load Balancers to the sub-10ms latencies that keep everything running smoothly.
Key Takeaways
- Application Load Balancer (ALB) can handle up to 1,000,000 requests per second under optimal conditions with sufficient backend capacity
- Network Load Balancer (NLB) supports up to 3,500,000 requests per second for TCP traffic with 100 Gbps bandwidth per load balancer
- Gateway Load Balancer (GNLB) processes up to 100 Gbps of traffic per load balancer with Geneve encapsulation
- ALB scales to 100 LCUs per load balancer automatically
- NLB supports up to 1,000 targets per target group across AZs
- Gateway Load Balancer Endpoint scales to 10 Gbps per endpoint
- ALB 99.99% SLA when deployed across 2+ AZs
- NLB delivers 99.995% availability over 30-day period
- Gateway Load Balancer maintains 99.99% uptime with GWLBE redundancy
- ALB priced at $0.0225 per hour + $0.008 per LCU-hour in us-east-1
- NLB costs $0.0225 per hour + $0.006 per million LCUs in us-east-1
- Gateway LB $0.0125 per hour + $0.004 per LCU-hour base rate
- ALB supports HTTP/HTTPS, gRPC, WebSocket, HTTP/2 protocols natively
- NLB handles TCP, UDP, TLS, TCP_UDP passthrough at L4
- Gateway LB integrates with third-party virtual appliances via GENEVE
Each AWS load balancer type offers distinct high performance and pricing strengths for different workloads.
Availability and Reliability
1ALB 99.99% SLA when deployed across 2+ AZs
Verified2NLB delivers 99.995% availability over 30-day period
Verified3Gateway Load Balancer maintains 99.99% uptime with GWLBE redundancy
Verified4Classic LB health checks detect failures in 15 seconds interval
Directional5ALB unhealthy target replacement automatic within 60 seconds
Single source6NLB supports DNS failover with Route 53 for 100% zonal isolation
Verified7Load balancer CloudWatch metric UnHealthyHostCount threshold 10% for alarms
Verified8ALB multi-AZ deployment survives single AZ outage with 0 downtime
Verified9NLB TCP connection resilience to backend failures 99.99%
Directional10Gateway LB transparent failover to healthy appliances 100% packet forwarding
Single source11Classic LB cross-zone balancing ensures no single point failure
Verified12ALB access logs capture 100% of requests for auditing reliability
Verified13NLB preserves flow during AZ failures with static IPs
Verified14Load balancer HTTP 5xx error rate <0.01% in steady state
Directional15ALB Lambda targets failover automatically to healthy functions
Single source16NLB health check grace period up to 3600 seconds post-launch
Verified17Gateway LB supports active/active appliance clusters for HA
Verified18Classic LB SSL negotiation failure rate 0.001% with policy updates
Verified19ALB target health state propagates in <30 seconds across regions
Directional20NLB UDP stateless handling ensures 100% delivery to targets
Single source21Load balancer deletion protection prevents accidental downtime 100%
Verified22ALB WAF rules enforce 99.9% DDoS mitigation availability
Verified23NLB Global Accelerator integration boosts availability to 99.999%
Verified24Gateway LB ECMP hashing for even flow distribution 99.99% reliability
Directional25Classic LB Surge Protector throttles to maintain 99.9% availability
Single source26ALB connection multiplexing sustains 1 million concurrent connections reliably
Verified27NLB preserves client-to-target affinity during failovers
Verified28Load balancer metrics retention 15 months in CloudWatch
Verified29ALB ALB-to-ALB networking latency <5ms intra-region
DirectionalAvailability and Reliability Interpretation
AWS load balancers offer a spectrum of reliably boring excellence, from the Classic’s dependable grunt work to the ALB’s intelligent orchestration, the NLB’s stoic resilience, and the Gateway’s seamless transparency, ensuring your application’s uptime is a foregone conclusion rather than a hopeful gamble.
Feature Specifications
1ALB supports HTTP/HTTPS, gRPC, WebSocket, HTTP/2 protocols natively
Verified2NLB handles TCP, UDP, TLS, TCP_UDP passthrough at L4
Verified3Gateway LB integrates with third-party virtual appliances via GENEVE
Verified4Classic LB supports EC2-Classic and VPC with sticky sessions
Directional5ALB content-based routing on host, path, headers, methods, query
Single source6NLB static/private IP per subnet for predictable routing
Verified7Load balancer IAM authentication for console/API access
Verified8ALB integrates with ACM for free public SSL certificates
Verified9NLB supports AWS PrivateLink for private connectivity
Directional10Gateway LB VLAN encapsulation for IDS/IPS/NFW appliances
Single source11Classic LB CloudWatch alarms on latency, request count, errors
Verified12ALB target types: IP, Instance, Lambda, ALB chaining
Verified13NLB client IP preservation and source IP transparency
Verified14Load balancer access logs in JSON format to S3
Directional15ALB WAF and Shield integration for security rules
Single source16NLB supports IPv4/IPv6 dual-stack endpoints
Verified17Gateway LB ECMP hashing algorithms for flow stickiness
Verified18Classic LB SSL offload with multiple security policies
Verified19ALB authentication with OIDC, SAML, Cognito, Lambda auth
Directional20NLB integrates with Route 53 for DNS health checks
Single source21Load balancer supports Terraform, CloudFormation, CDK IaC
Verified22ALB redirect/return fixed responses 3xx actions
Verified23NLB TLS termination with SNI for multiple certs
Verified24Gateway LB appliance health monitoring via metadata
Directional25Classic LB app cookie stickiness duration-based
Single source26ALB weighted target groups for blue-green deployments
Verified27NLB cross-zone disabled by default for zonal resilience
VerifiedFeature Specifications Interpretation
Think of Elastic Load Balancing as a family where each member has a distinct personality: ALB is the discerning, multilingual, and security-conscious concierge; NLB is the no-nonsense, high-performance traffic director; Gateway LB is the specialist who bridges worlds with custom tunnels; and Classic LB is the reliable, straightforward relative you can still count on for sticky, old-school solutions.
Performance Metrics
1Application Load Balancer (ALB) can handle up to 1,000,000 requests per second under optimal conditions with sufficient backend capacity
Verified2Network Load Balancer (NLB) supports up to 3,500,000 requests per second for TCP traffic with 100 Gbps bandwidth per load balancer
Verified3Gateway Load Balancer (GNLB) processes up to 100 Gbps of traffic per load balancer with Geneve encapsulation
Verified4Classic Load Balancer achieves average latency of under 10ms for HTTP requests in us-east-1 region
Directional5ALB TLS termination latency is typically 1-2ms per connection with AWS-2010 certificate
Single source6NLB preserves client IP with UDP traffic handling up to 400,000 connections per second
Verified7ALB supports WebSocket connections with throughput of 10,000 messages per second per target
Verified8Load balancer CloudWatch metric HealthyHostCount averages 99.9% across global deployments
Verified9ALB HTTP/2 support reduces latency by up to 30% compared to HTTP/1.1 multiplexing
Directional10NLB TCP idle timeout is configurable up to 4000 seconds, minimizing reconnections
Single source11ALB processes 1 LCU equivalent to max(25 new connections/sec, 3,000 active connections, 1 GB processed/sec, 2,500 fast connections/sec)
Verified12Gateway Load Balancer handles 10,000 appliances per GWLB endpoint
Verified13Classic LB supports up to 10,000 certificates per load balancer for SSL termination
Verified14ALB target response time metric p99 is under 500ms in production benchmarks
Directional15NLB supports 1 million DNS queries per second resolution time under 50ms
Single source16ALB Lambda integration processes 1,000 invocations per second per target group
Verified17Load balancer RequestCount metric spikes to 500,000/sec during peak loads without degradation
Verified18NLB Zonal isolation ensures 99.99% intra-zone latency under 1ms
Verified19ALB gRPC support handles 50,000 streams per second per load balancer
Directional20Classic LB SurgeQueueLength max 1,024 pending requests before 503 errors
Single source21ALB HTTP code 200 response rate 99.5% in standard workloads
Verified22NLB UDP packets processed at 1 million per second per port
Verified23Gateway LB IPsec VPN throughput 25 Gbps aggregate
Verified24ALB connection balancing distributes evenly across 1,000 targets with <1% variance
Directional25Load balancer ElapsedTime p50 median 50ms for ALB HTTP requests
Single source26NLB TLS 1.3 handshake time average 20ms
Verified27ALB WAF integration blocks 99.99% of SQLi attacks with <5ms overhead
Verified28Classic LB backend connection reuse rate 95% in persistent mode
Verified29NLB preserve source IP for 100% of TCP/UDP flows accurately
Directional30ALB fixed-response actions process 10,000 redirects/sec without backend load
Single sourcePerformance Metrics Interpretation
Amazon's load balancers are like a well-orchestrated, if slightly obsessive, symphony where each instrument—whether it's the ALB's virtuosic request handling, the NLB's relentless packet-pushing, or the GWLB's steadfast traffic-shepherding—is precisely tuned to ensure that the internet's cacophony arrives at your doorstep not as a chaotic crash, but as a polite, sub-500ms knock.
Pricing Details
1ALB priced at $0.0225 per hour + $0.008 per LCU-hour in us-east-1
Verified2NLB costs $0.0225 per hour + $0.006 per million LCUs in us-east-1
Verified3Gateway LB $0.0125 per hour + $0.004 per LCU-hour base rate
Verified4Classic LB $0.025 per hour + $0.008 per GB data processed
Directional5ALB LCU calculation: 1 LCU = 25 connections/sec or 3k active or 1GB/sec or 2.5k/sec fast
Single source6NLB hourly charge waived first 15 min during creation
Verified7Load balancer IPv6 free additional to IPv4 charges
Verified8ALB data transfer out to internet $0.008/GB first 10TB/month
Verified9NLB elastic IP hourly $0.005 beyond one per LB
Directional10Gateway LB no charge for hours with zero LCUs processed
Single source11Classic LB no charge for backend data transfer within same AZ
Verified12ALB Savings Plans discount up to 72% off on-demand pricing
Verified13NLB 1-year Reserved Instance 40% savings
Verified14Load balancer Free Tier 750 hours/month first year per account
Directional15ALB WAF association $5/rule/month + $1/million requests
Single source16NLB Global Accelerator passthrough pricing $0.025/GB
Verified17Gateway LB VPC endpoint $0.01/hour per endpoint
Verified18Classic LB certificate import free, ACM integration free
Verified19ALB access logs to S3 standard storage $0.023/GB/month
Directional20NLB CloudWatch detailed monitoring $0.30/million metrics
Single source21Load balancer tag-based billing granularity per resource
Verified22ALB regional pricing variance 10% lower in eu-west-1 vs us-east-1
Verified23NLB LCU-hours billed per second minimum 1 minute
Verified24Gateway LB no ingress/egress data fees within VPC
Directional25Classic LB cross-AZ data $0.01/GB transferred
Single source26ALB Spot Instances integration no premium pricing
VerifiedPricing Details Interpretation
Balancing your cloud budget requires as much precision as your architecture, because these layered and conditional fees—from per-second LCU charges and waived introductory windows to regional variances and Savings Plans—create a financial puzzle where your load balancer's operational elegance is directly rivaled by the complexity of its invoice.
Scalability Limits
1ALB scales to 100 LCUs per load balancer automatically
Verified2NLB supports up to 1,000 targets per target group across AZs
Verified3Gateway Load Balancer Endpoint scales to 10 Gbps per endpoint
Verified4Classic LB maximum 20 listeners per load balancer
Directional5ALB allows 100 rules per ALB listener for routing
Single source6NLB target groups limited to 3,000 targets total per NLB
Verified7Load balancer quota 20 ALBs per region default, increasable to 100
Verified8ALB deregistration delay max 3600 seconds for graceful shutdowns
Verified9NLB supports 200 subnets per NLB for zonal deployment
Directional10Gateway LB appliances up to 100 per GWLBE
Single source11Classic LB maximum 100 backend instances per port
Verified12ALB host-based routing supports 100 host headers per rule set
Verified13NLB IP targets up to 200 IPv4/IPv6 per target group
Verified14Load balancer attributes configurable up to 50 custom attributes per LB
Directional15ALB path-based routing patterns up to 5 per rule with wildcards
Single source16NLB cross-zone load balancing supports unlimited AZs with even distribution
Verified17Gateway LB VLAN mode supports 65,536 MAC addresses per endpoint
Verified18Classic LB VPC maximum 20 LBs per VPC
Verified19ALB query string parameters match up to 10 per condition
Directional20NLB static IP mode 1 IP per AZ per subnet
Single source21Load balancer access logs retention up to 365 days in S3
Verified22ALB sticky sessions cookie duration max 7 days
Verified23NLB health checks interval min 10 seconds, max 300 seconds
Verified24Gateway LB MTU 8500 bytes max for jumbo frames
Directional25ALB slow start duration max 900 seconds per target
Single source26Classic LB connection draining timeout 20 minutes max
Verified27NLB deregistration delay 300 seconds default
Verified28ALB source IP CIDR blocks up to 64 per rule
Verified29Load balancer tags max 50 per resource
Directional30NLB client IP preservation mode for all protocols up to 1 million concurrent
Single sourceScalability Limits Interpretation
While each load balancer generously defines its own unique boundaries for scalability, from the ALB's 100-rule finesse to the NLB's brute-force million-connection endurance, they collectively form a meticulous and sometimes quirky rulebook for architecting robust traffic flow in the cloud.
